1st Place · TSA Nationals

Loading slides…
PythonGISArcGISSpatial AnalysisDisaster Response

Fault Lines & Front Lines is a geospatial analysis portfolio we built for TSA's (Technology Student Association) Geospatial Technology event, where the 2024–2025 theme asked teams to identify a disaster threat facing their own community. We're in King County, Washington, so picking earthquakes was not a hard decision: the county sits in the Pacific Ring of Fire, in the Puget Sound region, near or directly on top of the Seattle Fault, the Tacoma Fault, and the Cascadia Subduction Zone, with active volcanoes like Mount Rainier and Mount Baker on either side. The 2001 magnitude-6.8 Nisqually earthquake is the most recent reminder of all that.

We pulled fault line, liquefaction, floodplain, bridge-condition, unreinforced masonry building, and emergency-services data from sources like the USGS Earthquake Catalog, King County GIS Open Data, Seattle GeoData, and Esri's federal fault datasets, then overlaid it across eight analysis maps to find Seattle's highest-risk zones and inform evacuation and shelter-in-place guidance.

Seattle is the most populous city in Washington, and its residents face outsized risk in a major earthquake. Dense downtown neighborhoods sit close to the Seattle Fault, over a thousand unreinforced masonry (URM) buildings predate 1945 and were never secured to code, and roughly 2,000 miles of underground wastewater pipeline crisscross known fault zones. A map of where earthquakes have historically been recorded doesn't show any of that. Emergency planning needs to see where fault activity, unstable soil, vulnerable buildings, and critical infrastructure actually overlap.

  • Source data came from a bunch of different public portals (USGS, King County GIS, Seattle GeoData, FEMA, Esri) in inconsistent formats with no shared schema.
  • The Seattle population density dataset alone had ~27,000 census-block records, which is way too many to render as one readable map.
  • Judges and the public mostly aren't GIS people, so the findings had to work as a written analysis and an infographic on top of the raw maps.
  • Nobody can predict the location, timing, or magnitude of an earthquake, so any "prediction" we made had to be framed as a probabilistic estimate and not a forecast.

Averaging historical epicenters to model a plausible future earthquake

Instead of guessing at a location, we used NumPy to average the epicenter coordinates of every magnitude 2.5+ earthquake recorded in the area from 1900 to 2025 (the data behind Figure 1) and treated that average as a probable future epicenter. Then we used the empirical relationship Area (km²) = 10^(0.5M − 1.8) to size the impact radius of a predicted M7.0 around it, and overlaid that on the URM building locations.

Reason:
Averaging real historical activity is a defensible, reproducible way to pick a hypothetical epicenter, which beats dropping a pin on downtown because it looks dramatic.

Tradeoff:
Averaging treats every historical earthquake as equally informative no matter its magnitude or how recent it was, so the point you get is the statistical center of past activity, not a real seismological forecast.

ETAS modeling for aftershock intensity, converted to Modified Mercalli Intensity

We implemented an Epidemic-Type Aftershock Sequence (ETAS) model in Python/NumPy to estimate where and when aftershocks would hit after a simulated M7.0, converted the resulting log-scale intensity into Peak Ground Acceleration, and then converted that into the Modified Mercalli Intensity (MMI) scale so non-scientists could read it.

Reason:
Aftershock risk usually gets reported in scientific units (PGA, log-intensity) that don't mean much to city planners or the public. MMI is the scale Seattle's own hazard planning documents use, so converting to it made the output directly usable.

Tradeoff:
The conversion uses an approximate PGA-to-MMI regression instead of region-specific attenuation data, so the absolute MMI values are indicative, not precise.

Composite hazard-per-structure overlay instead of single-factor maps

Analysis Map #2 stacks liquefaction-prone areas, FEMA floodplain boundaries, and an average-hazards-per-structure grid on top of each other instead of giving each hazard its own map.

Reason:
Liquefaction, flooding, and structural risk compound each other in the same low-lying neighborhoods (South and West Seattle, for example). Showing them separately hides that overlap, and the overlap is exactly where mitigation resources should go first.

Tradeoff:
Stacking multiple choropleth layers makes the map denser and harder to read at a glance than a single-variable heatmap.

  • Won 1st place in Geospatial Technology at the Washington TSA State Conference, which qualified us for TSA Nationals in Nashville, Tennessee.
  • Made and cited 8 original analysis maps covering fault lines, bridge seismic risk, liquefaction/flood overlays, unreinforced masonry buildings, emergency services, wastewater infrastructure, and aftershock modeling.
  • Built Python/NumPy models (mean-epicenter estimation, ETAS aftershock intensity, PGA→MMI conversion) to get from raw historical earthquake data to a predicted-impact map a normal person can read.
  • Concluded that a magnitude 6+ earthquake on the Seattle Fault could cause at least $50 billion in damage. For comparison, the 2001 Nisqually earthquake did ~$4 billion near Olympia, which is much smaller than Seattle, with its ~755,000 people and heavy concentration of tech-sector jobs.

  • South and West Seattle only really stand out once liquefaction, flooding, and structural risk are on the same map. Any one of those layers on its own undersells them.
  • After all the ETAS modeling, our most actionable conclusion was basically "shelter in place, don't evacuate over an unstable bridge," which is a little funny, but it's also the one thing a resident can actually do something with.
  • Public GIS data (USGS, county open-data portals, FEMA) is split across a lot of sources in different formats, and reconciling it was most of the real work.
  • You can't predict an earthquake, but an averaged epicenter and an ETAS aftershock model are still worth building as long as you're honest that they're estimates.